San Francisco, CA – A new perk is rapidly gaining traction in Silicon Valley’s fierce competition for engineering talent: direct access to artificial intelligence compute, allocated as a budget of “AI tokens.” The concept, initially discussed among venture capitalists, was thrust into the spotlight this week after Nvidia CEO Jensen Huang proposed engineers receive an amount of tokens equivalent to half their base salary.
Huang, speaking at the company’s annual GPU Technology Conference, framed the tokens as a productivity multiplier. Engineers can spend these tokens – computational units used to power AI tools like Claude, ChatGPT, and Gemini – to run AI agents, automate tasks, and accelerate coding projects. “Every engineer that has access to tokens will be more productive,” Huang said, according to CNBC reporting. He estimated top engineers could utilize $250,000 worth of AI compute annually.
The idea of factoring in compute costs to engineering compensation wasn’t born at Nvidia. Tomasz Tunguz, a venture capitalist at Theory Ventures, wrote in February that tech startups were already beginning to treat inference costs as a “fourth component to engineering compensation.” Using data from Levels.fyi, Tunguz calculated that a top-quartile software engineer earning $375,000 could observe their total compensation package rise to $475,000 with an additional $100,000 in tokens.
The rise of “agentic AI” is driving the demand for tokens. Unlike traditional AI that responds to individual prompts, agentic AI systems – exemplified by the open-source assistant OpenClaw, released in January – operate autonomously, continuously executing tasks and spawning sub-agents. This shift has led to a surge in token consumption, with some engineers at companies like Meta and OpenAI reportedly competing on internal leaderboards tracking their usage, according to a recent report in the New York Times.
While presented as a benefit, experts caution that AI token budgets could reach with increased pressure on engineers. A substantial token allocation implies a corresponding expectation of increased output. Companies are, in effect, funding a significant amount of compute power on behalf of each engineer, creating an implicit demand for higher productivity.
Financial implications also loom. As token spending approaches or exceeds an employee’s salary, the fundamental economics of staffing begin to shift, raising questions about the optimal ratio of human engineers to automated compute. Jamaal Glenn, a former venture capital investor and current CFO, noted that token allowances could be a way for companies to inflate the perceived value of compensation packages without increasing cash or equity – components that traditionally build long-term wealth for employees. Tokens do not vest, appreciate, or factor into future salary negotiations.
The potential for companies to normalize tokens as a standard part of pay raises concerns about maintaining competitive cash compensation and equity grants. While a generous token budget may appear attractive, its long-term value remains uncertain. The rapid evolution of AI technology means that the value and utility of tokens could change quickly, leaving engineers reliant on a perk that doesn’t offer the same security as traditional compensation elements.
